A Hybrid DSP/Deep Learning Approach to Real-Time Full-Band Speech Enhancement

Jean-Marc Valin
9/24/2017
cs.SDeess.AS

Official code: xiph/rnnoise (5,904 stars, BSD-3-Clause, last commit 2025-02-22).

Abstract

Despite noise suppression being a mature area in signal processing, it remains highly dependent on fine tuning of estimator algorithms and parameters. In this paper, we demonstrate a hybrid DSP/deep learning approach to noise suppression. A deep neural network with four hidden layers is used to estimate ideal critical band gains, while a more traditional pitch filter attenuates noise between pitch harmonics. The approach achieves significantly higher quality than a traditional minimum mean squared error spectral estimator, while keeping the complexity low enough for real-time operation at 48 kHz on a low-power processor.

View on arXivView PDF

Code Implementations(1)

xiph/rnnoiseOfficial100%
BSD-3-Clause

Recurrent neural network for audio noise reduction

5,9041,082C, M4Aug 3, 20171 years agoBSD-3-Clause
audiocnoise-reductionrnn

Similar Papers

Effects of Dataset Sampling Rate for Noise Cancellation through Deep Learning

82% match2024

Net2Net: When Un-trained Meets Pre-trained Networks for Robust Real-World Denoising

81% match2025

Dilated CNNs for Periodic Signal Processing: A Low-Complexity Approach

79% match2026

A Hybrid SFANC-FxNLMS Algorithm for Active Noise Control based on Deep Learning

79% match1 repo862022

IS${}^3$ : Generic Impulsive--Stationary Sound Separation in Acoustic Scenes using Deep Filtering

79% match1 repo62025

Ultra Low Complexity Deep Learning Based Noise Suppression

79% match1 repo282023

Cite this paper

@article{valin2017hybrid,
  title  = {A Hybrid DSP/Deep Learning Approach to Real-Time Full-Band Speech Enhancement},
  author = {Jean-Marc Valin},
  year   = {2017},
  eprint = {1709.08243},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1709.08243}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1709.08243.svg)](https://codeofpaper.com/paper/1709.08243)

Discussion